1
0
mirror of https://github.com/mailcow/mailcow-dockerized.git synced 2026-07-25 06:03:33 +00:00

[Agent] Replace dockerapi container with Redis-based control bus

This commit is contained in:
FreddleSpl0it
2026-05-20 20:54:51 +02:00
parent 4ddcee28e4
commit 689d753264
75 changed files with 3740 additions and 2462 deletions
@@ -0,0 +1,9 @@
package services
import "time"
// nowStamp returns a sortable timestamp used to suffix moved/garbage maildirs
// so repeated cleanups don't collide.
func nowStamp() string {
return time.Now().UTC().Format("20060102T150405Z")
}
@@ -0,0 +1,294 @@
package services
import (
"context"
"fmt"
"net"
"os"
"path/filepath"
"strings"
"syscall"
"time"
"github.com/mailcow/mailcow-dockerized/agent/internal/commands"
"github.com/mailcow/mailcow-dockerized/agent/internal/proc"
)
func init() { Register("dovecot", buildDovecot) }
const vmailRoot = "/var/vmail"
func dovecotHealthProbe(ctx context.Context) error {
// IMAP greeting on :143 — must be "* OK ..."
conn, err := net.DialTimeout("tcp", "127.0.0.1:143", 3*time.Second)
if err != nil {
return err
}
defer conn.Close()
buf := make([]byte, 64)
_ = conn.SetReadDeadline(time.Now().Add(3 * time.Second))
n, err := conn.Read(buf)
if err != nil {
return fmt.Errorf("read greeting: %w", err)
}
greeting := string(buf[:n])
if !strings.HasPrefix(greeting, "* OK") {
return fmt.Errorf("unexpected greeting: %s", strings.TrimSpace(greeting))
}
return nil
}
func buildDovecot(sup *proc.Supervisor) *commands.Table {
t := commands.New("dovecot")
t.HealthProbe = dovecotHealthProbe
// `dovecot reload` re-reads config without restarting the master process.
t.Register("reload", func(ctx context.Context, _ map[string]any) (any, error) {
r, err := commands.Run(ctx, commands.RunOptions{}, "dovecot", "reload")
return nil, asError(r, err)
})
addLifecycleExceptReload(t, sup)
t.Register("exec.fts-rescan", func(ctx context.Context, args map[string]any) (any, error) {
user := commands.ArgStringOpt(args, "user", "")
argv := []string{"doveadm", "fts", "rescan"}
if user != "" {
argv = append(argv, "-u", user)
} else {
argv = append(argv, "-A")
}
r, err := commands.Run(ctx, commands.RunOptions{}, argv...)
return nil, asError(r, err)
})
t.Register("exec.sieve-list", func(ctx context.Context, args map[string]any) (any, error) {
user, err := commands.ArgString(args, "user")
if err != nil {
return nil, err
}
r, err := commands.Run(ctx, commands.RunOptions{}, "doveadm", "sieve", "list", "-u", user)
if err != nil {
return nil, err
}
if r.ExitCode != 0 {
return nil, &runError{msg: strings.TrimSpace(r.Stderr)}
}
scripts := splitNonEmpty(r.Stdout)
return map[string]any{"scripts": scripts}, nil
})
t.Register("exec.sieve-print", func(ctx context.Context, args map[string]any) (any, error) {
user, err := commands.ArgString(args, "user")
if err != nil {
return nil, err
}
script, err := commands.ArgString(args, "script")
if err != nil {
return nil, err
}
r, err := commands.Run(ctx, commands.RunOptions{OutputCap: 1 << 20}, "doveadm", "sieve", "get", "-u", user, script)
if err != nil {
return nil, err
}
if r.ExitCode != 0 {
return nil, &runError{msg: strings.TrimSpace(r.Stderr)}
}
return map[string]any{"body": r.Stdout}, nil
})
t.Register("exec.acl-get", func(ctx context.Context, args map[string]any) (any, error) {
user, err := commands.ArgString(args, "user")
if err != nil {
return nil, err
}
// First enumerate mailboxes, then collect ACLs per mailbox.
boxes, err := commands.Run(ctx, commands.RunOptions{}, "doveadm", "mailbox", "list", "-u", user)
if err != nil {
return nil, err
}
if boxes.ExitCode != 0 {
return nil, &runError{msg: strings.TrimSpace(boxes.Stderr)}
}
out := []map[string]any{}
for _, mbx := range splitNonEmpty(boxes.Stdout) {
r, err := commands.Run(ctx, commands.RunOptions{}, "doveadm", "acl", "get", "-u", user, mbx)
if err != nil || r.ExitCode != 0 {
continue
}
for _, line := range strings.Split(strings.TrimSpace(r.Stdout), "\n") {
line = strings.TrimSpace(line)
if line == "" || strings.HasPrefix(line, "ID") {
continue
}
fields := strings.Fields(line)
if len(fields) >= 2 {
out = append(out, map[string]any{
"mailbox": mbx,
"identifier": fields[0],
"rights": strings.Join(fields[1:], " "),
})
}
}
}
return map[string]any{"acls": out}, nil
})
t.Register("exec.acl-set", func(ctx context.Context, args map[string]any) (any, error) {
user, err := commands.ArgString(args, "user")
if err != nil {
return nil, err
}
mailbox, err := commands.ArgString(args, "mailbox")
if err != nil {
return nil, err
}
identifier, err := commands.ArgString(args, "identifier")
if err != nil {
return nil, err
}
rights, err := commands.ArgString(args, "rights")
if err != nil {
return nil, err
}
r, err := commands.Run(ctx, commands.RunOptions{}, "doveadm", "acl", "set", "-u", user, mailbox, identifier, rights)
return nil, asError(r, err)
})
t.Register("exec.acl-delete", func(ctx context.Context, args map[string]any) (any, error) {
user, err := commands.ArgString(args, "user")
if err != nil {
return nil, err
}
mailbox, err := commands.ArgString(args, "mailbox")
if err != nil {
return nil, err
}
identifier, err := commands.ArgString(args, "identifier")
if err != nil {
return nil, err
}
r, err := commands.Run(ctx, commands.RunOptions{}, "doveadm", "acl", "delete", "-u", user, mailbox, identifier)
return nil, asError(r, err)
})
t.Register("exec.maildir-cleanup", func(ctx context.Context, args map[string]any) (any, error) {
maildir, err := commands.ArgString(args, "maildir")
if err != nil {
return nil, err
}
if err := assertSafeMaildirPath(maildir); err != nil {
return nil, err
}
src := filepath.Join(vmailRoot, maildir)
dst := filepath.Join(vmailRoot, "_garbage", maildir+"_"+nowStamp())
if _, err := os.Stat(src); os.IsNotExist(err) {
return nil, commands.ErrNotFound
}
if err := os.MkdirAll(filepath.Dir(dst), 0o770); err != nil {
return nil, err
}
return nil, os.Rename(src, dst)
})
t.Register("exec.df", func(ctx context.Context, args map[string]any) (any, error) {
dir := commands.ArgStringOpt(args, "dir", "/var/vmail")
var st syscall.Statfs_t
if err := syscall.Statfs(dir, &st); err != nil {
return nil, err
}
size := uint64(st.Blocks) * uint64(st.Bsize)
free := uint64(st.Bavail) * uint64(st.Bsize)
used := size - free
pct := 0
if size > 0 {
pct = int(float64(used) / float64(size) * 100)
}
// Format: Filesystem,Size,Used,Avail,Use%,Mounted-on
return fmt.Sprintf("%s,%s,%s,%s,%d%%,%s",
"local", humanBytes(size), humanBytes(used), humanBytes(free), pct, dir), nil
})
t.Register("exec.maildir-move", func(ctx context.Context, args map[string]any) (any, error) {
from, err := commands.ArgString(args, "from")
if err != nil {
return nil, err
}
to, err := commands.ArgString(args, "to")
if err != nil {
return nil, err
}
if err := assertSafeMaildirPath(from); err != nil {
return nil, err
}
if err := assertSafeMaildirPath(to); err != nil {
return nil, err
}
src := filepath.Join(vmailRoot, from)
dst := filepath.Join(vmailRoot, to)
if _, err := os.Stat(src); os.IsNotExist(err) {
return nil, commands.ErrNotFound
}
if err := os.MkdirAll(filepath.Dir(dst), 0o770); err != nil {
return nil, err
}
return nil, os.Rename(src, dst)
})
return t
}
// addLifecycleExceptReload wires restart/stop/start without overriding reload,
// which postfix/dovecot define themselves (canonical CLI command).
func addLifecycleExceptReload(t *commands.Table, sup *proc.Supervisor) {
if sup == nil {
return
}
t.Register("restart", func(ctx context.Context, _ map[string]any) (any, error) {
return nil, sup.Restart(ctx)
})
t.Register("stop", func(ctx context.Context, _ map[string]any) (any, error) {
return nil, sup.Stop(ctx)
})
t.Register("start", func(ctx context.Context, _ map[string]any) (any, error) {
return nil, sup.Start()
})
}
func splitNonEmpty(s string) []string {
out := []string{}
for _, line := range strings.Split(strings.TrimSpace(s), "\n") {
line = strings.TrimSpace(line)
if line != "" {
out = append(out, line)
}
}
return out
}
// assertSafeMaildirPath blocks path traversal and absolute paths — relative
// names under /var/vmail only.
func assertSafeMaildirPath(p string) error {
if p == "" || strings.HasPrefix(p, "/") || strings.Contains(p, "..") {
return &validationErr{msg: "unsafe maildir path"}
}
return nil
}
type validationErr struct{ msg string }
func (e *validationErr) Error() string { return e.msg }
func (e *validationErr) Is(target error) bool { return target == commands.ErrValidation }
// humanBytes renders a byte count in `df -H` style (1000-based units).
func humanBytes(n uint64) string {
const unit = 1000
if n < unit {
return fmt.Sprintf("%dB", n)
}
div, exp := uint64(unit), 0
for x := n / unit; x >= unit; x /= unit {
div *= unit
exp++
}
return fmt.Sprintf("%.1f%c", float64(n)/float64(div), "KMGTPE"[exp])
}
@@ -0,0 +1,37 @@
package services
import (
"context"
"time"
"github.com/mailcow/mailcow-dockerized/agent/internal/commands"
"github.com/mailcow/mailcow-dockerized/agent/internal/proc"
)
// Services without any exec.* commands of their own — lifecycle only.
func init() {
Register("clamd", genericBuilder("clamd", tcpProbe("127.0.0.1:3310", 2*time.Second)))
Register("olefy", genericBuilder("olefy", tcpProbe("127.0.0.1:10055", 2*time.Second)))
Register("postfix-tlspol", genericBuilder("postfix-tlspol", tcpProbe("127.0.0.1:8642", 2*time.Second)))
Register("php-fpm", genericBuilder("php-fpm", tcpProbe("127.0.0.1:9001", 2*time.Second)))
Register("acme", genericBuilder("acme", nil))
Register("watchdog", genericBuilder("watchdog", nil))
Register("netfilter", genericBuilder("netfilter", nil))
Register("ofelia", genericBuilder("ofelia", nil))
Register("dovecot-fts", genericBuilder("dovecot-fts", nil))
}
func genericBuilder(name string, probe commands.HealthProbe) Builder {
return func(sup *proc.Supervisor) *commands.Table {
t := commands.New(name)
t.HealthProbe = probe
addLifecycle(t, sup)
return t
}
}
func tcpProbe(addr string, timeout time.Duration) commands.HealthProbe {
return func(ctx context.Context) error {
return probeTCP(addr, timeout)
}
}
@@ -0,0 +1,79 @@
package services
import (
"strings"
"github.com/mailcow/mailcow-dockerized/agent/internal/commands"
)
// runError is what we return when a shell command exited non-zero but the
// failure is not a "target not found" case. The bus maps it to
// ErrCodeInternal.
type runError struct{ msg string }
func (e *runError) Error() string { return e.msg }
// asError converts a (RunResult, err) pair from commands.Run into a single
// error: pre-exec error → return as-is; non-zero exit → wrap stderr.
func asError(r *commands.RunResult, err error) error {
if err != nil {
return err
}
if r.ExitCode != 0 {
msg := strings.TrimSpace(r.Stderr)
if msg == "" {
msg = "command exited " + itoa(r.ExitCode)
}
return &runError{msg: msg}
}
return nil
}
// asNotFoundOrError is the variant for queue/mailbox operations that may fail
// because the target doesn't live on this node. Maps known stderr fragments
// to commands.ErrNotFound so broadcast aggregation works.
func asNotFoundOrError(r *commands.RunResult, err error) error {
if err != nil {
return err
}
if r.ExitCode == 0 {
return nil
}
if matchesAny(r.Stderr, notFoundFragments) {
return commands.ErrNotFound
}
return &runError{msg: strings.TrimSpace(r.Stderr)}
}
func matchesAny(haystack string, fragments []string) bool {
for _, f := range fragments {
if strings.Contains(haystack, f) {
return true
}
}
return false
}
func itoa(i int) string {
// avoid strconv import for a one-shot; small ints only
if i == 0 {
return "0"
}
neg := false
if i < 0 {
neg = true
i = -i
}
var b [20]byte
n := len(b)
for i > 0 {
n--
b[n] = byte('0' + i%10)
i /= 10
}
if neg {
n--
b[n] = '-'
}
return string(b[n:])
}
@@ -0,0 +1,236 @@
package services
import (
"bufio"
"context"
"fmt"
"os"
"strconv"
"strings"
"syscall"
"time"
"github.com/mailcow/mailcow-dockerized/agent/internal/commands"
"github.com/mailcow/mailcow-dockerized/agent/internal/proc"
)
func init() { Register("host", buildHost) }
// hostProcRoot is where the host-agent container mounts /proc. If we're not
// running as host-agent, falling back to /proc still produces sensible numbers
// (the container's own view) so dashboards don't blank out in unit tests.
var hostProcRoot = "/host/proc"
func resolveProc(p string) string {
if _, err := os.Stat(hostProcRoot); err == nil {
return hostProcRoot + p
}
return "/proc" + p
}
func buildHost(_ *proc.Supervisor) *commands.Table {
t := commands.New("host")
// No lifecycle — the host-agent container has no main process to manage.
t.Register("exec.df", func(ctx context.Context, args map[string]any) (any, error) {
path := commands.ArgStringOpt(args, "path", "/")
var stat syscall.Statfs_t
if err := syscall.Statfs(path, &stat); err != nil {
return nil, fmt.Errorf("statfs %s: %w", path, err)
}
size := int64(stat.Blocks) * int64(stat.Bsize)
free := int64(stat.Bavail) * int64(stat.Bsize)
used := size - free
return map[string]any{
"path": path,
"size": size,
"used": used,
"available": free,
}, nil
})
t.Register("exec.host-stats", func(ctx context.Context, _ map[string]any) (any, error) {
return readHostStats()
})
return t
}
func readHostStats() (map[string]any, error) {
out := map[string]any{
"system_time": time.Now().Format("2006-01-02 15:04:05"),
"architecture": readArchitecture(),
}
if uptime, err := readUptime(); err == nil {
out["uptime"] = int64(uptime)
} else {
out["uptime"] = int64(0)
}
cores := readCPUCores()
cpuUsage, _ := sampleHostCPU(500 * time.Millisecond)
out["cpu"] = map[string]any{
"cores": cores,
"usage": cpuUsage,
}
memTotal, memUsage := readMemoryTotalAndUsagePct()
out["memory"] = map[string]any{
"total": memTotal, // bytes
"usage": memUsage, // percent 0..100
}
return out, nil
}
// readArchitecture returns the host's machine architecture (e.g. "x86_64",
// "aarch64"). Falls back to a single dash if syscall.Uname fails.
func readArchitecture() string {
var u syscall.Utsname
if err := syscall.Uname(&u); err != nil {
return "-"
}
return charsToString(u.Machine[:])
}
func charsToString(b []int8) string {
out := make([]byte, 0, len(b))
for _, c := range b {
if c == 0 {
break
}
out = append(out, byte(c))
}
return string(out)
}
// readCPUCores counts `^processor` lines in /proc/cpuinfo. On a container
// with /host/proc bind-mounted this gives the host's logical CPU count,
// not the container's cgroup limits.
func readCPUCores() int {
f, err := os.Open(resolveProc("/cpuinfo"))
if err != nil {
return 0
}
defer f.Close()
n := 0
sc := bufio.NewScanner(f)
for sc.Scan() {
if strings.HasPrefix(sc.Text(), "processor") {
n++
}
}
return n
}
// readMemoryTotalAndUsagePct reads /proc/meminfo and returns (total_bytes,
// usage_pct_0_100). "Usage" is computed as (Total - Available)/Total which
// matches what tools like `free` show as "used".
func readMemoryTotalAndUsagePct() (int64, int) {
f, err := os.Open(resolveProc("/meminfo"))
if err != nil {
return 0, 0
}
defer f.Close()
var total, available int64
sc := bufio.NewScanner(f)
for sc.Scan() {
fields := strings.Fields(sc.Text())
if len(fields) < 2 {
continue
}
switch fields[0] {
case "MemTotal:":
total = parseInt64(fields[1]) * 1024
case "MemAvailable:":
available = parseInt64(fields[1]) * 1024
}
}
if total <= 0 {
return 0, 0
}
used := total - available
if available <= 0 {
used = total
}
pct := int(float64(used) / float64(total) * 100.0)
if pct < 0 {
pct = 0
}
if pct > 100 {
pct = 100
}
return total, pct
}
func readUptime() (float64, error) {
b, err := os.ReadFile(resolveProc("/uptime"))
if err != nil {
return 0, err
}
fields := strings.Fields(string(b))
if len(fields) < 1 {
return 0, fmt.Errorf("malformed uptime")
}
return strconv.ParseFloat(fields[0], 64)
}
// sampleHostCPU returns CPU utilization (0..100) sampled over `window`.
func sampleHostCPU(window time.Duration) (float64, error) {
a, err := readCPULine()
if err != nil {
return 0, err
}
time.Sleep(window)
b, err := readCPULine()
if err != nil {
return 0, err
}
totalA, totalB := sum(a), sum(b)
idleA, idleB := a[3], b[3]
dTotal, dIdle := totalB-totalA, idleB-idleA
if dTotal == 0 {
return 0, nil
}
return 100.0 * float64(dTotal-dIdle) / float64(dTotal), nil
}
func readCPULine() ([]int64, error) {
f, err := os.Open(resolveProc("/stat"))
if err != nil {
return nil, err
}
defer f.Close()
sc := bufio.NewScanner(f)
if !sc.Scan() {
return nil, fmt.Errorf("empty /proc/stat")
}
fields := strings.Fields(sc.Text())
if len(fields) < 5 || fields[0] != "cpu" {
return nil, fmt.Errorf("unexpected /proc/stat first line")
}
out := make([]int64, 0, len(fields)-1)
for _, f := range fields[1:] {
n, err := strconv.ParseInt(f, 10, 64)
if err != nil {
return nil, err
}
out = append(out, n)
}
return out, nil
}
func sum(xs []int64) int64 {
var s int64
for _, x := range xs {
s += x
}
return s
}
func parseInt64(s string) int64 {
n, _ := strconv.ParseInt(s, 10, 64)
return n
}
@@ -0,0 +1,39 @@
package services
import (
"context"
"time"
"github.com/mailcow/mailcow-dockerized/agent/internal/commands"
"github.com/mailcow/mailcow-dockerized/agent/internal/proc"
)
func init() { Register("nginx", buildNginx) }
func nginxHealthProbe(ctx context.Context) error {
if err := probeShell(ctx, 3*time.Second, "nginx", "-t"); err != nil {
return err
}
return probeTCP("127.0.0.1:8081", 2*time.Second)
}
func buildNginx(sup *proc.Supervisor) *commands.Table {
t := commands.New("nginx")
t.HealthProbe = nginxHealthProbe
t.Register("reload", func(ctx context.Context, _ map[string]any) (any, error) {
r, err := commands.Run(ctx, commands.RunOptions{}, "nginx", "-s", "reload")
return nil, asError(r, err)
})
addLifecycleExceptReload(t, sup)
t.Register("exec.test-config", func(ctx context.Context, _ map[string]any) (any, error) {
r, err := commands.Run(ctx, commands.RunOptions{}, "nginx", "-t")
if err != nil {
return nil, err
}
return map[string]any{
"ok": r.ExitCode == 0,
"output": r.Stderr + r.Stdout,
}, nil
})
return t
}
@@ -0,0 +1,142 @@
package services
import (
"context"
"encoding/json"
"strings"
"time"
"github.com/mailcow/mailcow-dockerized/agent/internal/commands"
"github.com/mailcow/mailcow-dockerized/agent/internal/proc"
)
func init() { Register("postfix", buildPostfix) }
// notFoundFragments are substrings emitted by postsuper/postqueue when the
// requested queue id doesn't live on this node. Broadcast handlers map them
// to commands.ErrNotFound so the backend can count partial success.
var notFoundFragments = []string{
"No such file or directory",
"no such file",
"unknown",
}
func postfixHealthProbe(ctx context.Context) error {
if err := probeSMTPGreeting("127.0.0.1:25", 3*time.Second); err != nil {
return err
}
return probeShell(ctx, 5*time.Second, "postfix", "status")
}
func buildPostfix(sup *proc.Supervisor) *commands.Table {
t := commands.New("postfix")
t.HealthProbe = postfixHealthProbe
// Override generic reload — `postfix reload` is the canonical operation,
// not SIGHUP-to-supervisord (which would just rotate logs).
t.Register("reload", func(ctx context.Context, _ map[string]any) (any, error) {
r, err := commands.Run(ctx, commands.RunOptions{}, "postfix", "reload")
return nil, asError(r, err)
})
// Lifecycle: stop/start/restart still go through the supervisor.
if sup != nil {
t.Register("restart", func(ctx context.Context, _ map[string]any) (any, error) {
return nil, sup.Restart(ctx)
})
t.Register("stop", func(ctx context.Context, _ map[string]any) (any, error) {
return nil, sup.Stop(ctx)
})
t.Register("start", func(ctx context.Context, _ map[string]any) (any, error) {
return nil, sup.Start()
})
}
t.Register("exec.mailq", func(ctx context.Context, _ map[string]any) (any, error) {
r, err := commands.Run(ctx, commands.RunOptions{OutputCap: 8 << 20}, "postqueue", "-j")
if err != nil {
return nil, err
}
if r.ExitCode != 0 {
return nil, &runError{msg: "postqueue failed: " + r.Stderr}
}
// postqueue -j prints one JSON object per line.
entries := make([]map[string]any, 0)
for _, line := range strings.Split(strings.TrimSpace(r.Stdout), "\n") {
line = strings.TrimSpace(line)
if line == "" {
continue
}
var obj map[string]any
if err := json.Unmarshal([]byte(line), &obj); err == nil {
entries = append(entries, obj)
}
}
return map[string]any{"queue": entries}, nil
})
t.Register("exec.flush-queue", func(ctx context.Context, _ map[string]any) (any, error) {
r, err := commands.Run(ctx, commands.RunOptions{}, "postqueue", "-f")
return nil, asError(r, err)
})
t.Register("exec.delete-from-queue", func(ctx context.Context, args map[string]any) (any, error) {
qid, err := commands.ArgString(args, "queue_id")
if err != nil {
return nil, err
}
r, err := commands.Run(ctx, commands.RunOptions{}, "postsuper", "-d", qid)
return nil, asNotFoundOrError(r, err)
})
t.Register("exec.hold-queue", func(ctx context.Context, args map[string]any) (any, error) {
qid, err := commands.ArgString(args, "queue_id")
if err != nil {
return nil, err
}
r, err := commands.Run(ctx, commands.RunOptions{}, "postsuper", "-h", qid)
return nil, asNotFoundOrError(r, err)
})
t.Register("exec.unhold-queue", func(ctx context.Context, args map[string]any) (any, error) {
qid, err := commands.ArgString(args, "queue_id")
if err != nil {
return nil, err
}
r, err := commands.Run(ctx, commands.RunOptions{}, "postsuper", "-H", qid)
return nil, asNotFoundOrError(r, err)
})
t.Register("exec.deliver-now", func(ctx context.Context, args map[string]any) (any, error) {
qid, err := commands.ArgString(args, "queue_id")
if err != nil {
return nil, err
}
r, err := commands.Run(ctx, commands.RunOptions{}, "postqueue", "-i", qid)
return nil, asNotFoundOrError(r, err)
})
t.Register("exec.cat-queue", func(ctx context.Context, args map[string]any) (any, error) {
qid, err := commands.ArgString(args, "queue_id")
if err != nil {
return nil, err
}
r, err := commands.Run(ctx, commands.RunOptions{OutputCap: 2 << 20}, "postcat", "-q", qid)
if err != nil {
return nil, err
}
if r.ExitCode != 0 {
if matchesAny(r.Stderr, notFoundFragments) {
return nil, commands.ErrNotFound
}
return nil, &runError{msg: "postcat failed: " + r.Stderr}
}
return map[string]any{"body": r.Stdout}, nil
})
t.Register("exec.super-delete", func(ctx context.Context, _ map[string]any) (any, error) {
r, err := commands.Run(ctx, commands.RunOptions{}, "postsuper", "-d", "ALL")
return nil, asError(r, err)
})
return t
}
@@ -0,0 +1,81 @@
package services
import (
"bufio"
"context"
"errors"
"fmt"
"net"
"net/http"
"strings"
"time"
"github.com/mailcow/mailcow-dockerized/agent/internal/commands"
)
// probeTCP opens a TCP connection to addr within timeout. Returns nil if the
// port accepts a connection, otherwise the dial error.
func probeTCP(addr string, timeout time.Duration) error {
conn, err := net.DialTimeout("tcp", addr, timeout)
if err != nil {
return err
}
_ = conn.Close()
return nil
}
// probeSMTPGreeting connects to addr and reads the SMTP greeting line. The
// service is considered healthy if the line starts with "220".
func probeSMTPGreeting(addr string, timeout time.Duration) error {
conn, err := net.DialTimeout("tcp", addr, timeout)
if err != nil {
return err
}
defer conn.Close()
_ = conn.SetReadDeadline(time.Now().Add(timeout))
line, err := bufio.NewReader(conn).ReadString('\n')
if err != nil {
return fmt.Errorf("read greeting: %w", err)
}
if !strings.HasPrefix(line, "220") {
return fmt.Errorf("unexpected greeting: %s", strings.TrimSpace(line))
}
return nil
}
// probeHTTP issues a GET to url, checks for a 2xx status.
func probeHTTP(ctx context.Context, url string, timeout time.Duration) error {
cctx, cancel := context.WithTimeout(ctx, timeout)
defer cancel()
req, err := http.NewRequestWithContext(cctx, http.MethodGet, url, nil)
if err != nil {
return err
}
resp, err := http.DefaultClient.Do(req)
if err != nil {
return err
}
defer resp.Body.Close()
if resp.StatusCode < 200 || resp.StatusCode >= 300 {
return fmt.Errorf("http %s", resp.Status)
}
return nil
}
// probeShell runs argv with a timeout and returns nil if exit code is 0.
func probeShell(ctx context.Context, timeout time.Duration, argv ...string) error {
cctx, cancel := context.WithTimeout(ctx, timeout)
defer cancel()
r, err := commands.Run(cctx, commands.RunOptions{}, argv...)
if err != nil {
return err
}
if r.ExitCode != 0 {
msg := strings.TrimSpace(r.Stderr)
if msg == "" {
msg = fmt.Sprintf("exit %d", r.ExitCode)
}
return errors.New(msg)
}
return nil
}
@@ -0,0 +1,86 @@
package services
import (
"context"
"os"
"path/filepath"
"strings"
"time"
"github.com/mailcow/mailcow-dockerized/agent/internal/commands"
"github.com/mailcow/mailcow-dockerized/agent/internal/proc"
)
func init() { Register("rspamd", buildRspamd) }
func rspamdHealthProbe(ctx context.Context) error {
return probeHTTP(ctx, "http://127.0.0.1:11334/ping", 3*time.Second)
}
// Override file rspamd reads on startup for the controller's enable_password.
const rspamdWorkerPasswordPath = "/etc/rspamd/override.d/worker-controller-password.inc"
func buildRspamd(sup *proc.Supervisor) *commands.Table {
t := commands.New("rspamd")
t.HealthProbe = rspamdHealthProbe
addLifecycle(t, sup)
t.Register("exec.set-worker-password", func(ctx context.Context, args map[string]any) (any, error) {
password, err := commands.ArgString(args, "password")
if err != nil {
return nil, err
}
// rspamadm pw -e -p <pw> writes the hashed value to stdout.
r, err := commands.Run(ctx, commands.RunOptions{}, "rspamadm", "pw", "-e", "-p", password)
if err != nil {
return nil, err
}
if r.ExitCode != 0 {
return nil, &runError{msg: "rspamadm pw failed: " + strings.TrimSpace(r.Stderr)}
}
hash := strings.TrimSpace(r.Stdout)
// rspamd distinguishes `password` (read-only access to the controller)
// from `enable_password` (write access — restart, settings, learn).
content := "enable_password = \"" + hash + "\";\n"
if err := os.MkdirAll(filepath.Dir(rspamdWorkerPasswordPath), 0o755); err != nil {
return nil, err
}
if err := os.WriteFile(rspamdWorkerPasswordPath, []byte(content), 0o644); err != nil {
return nil, err
}
// Must do a full re-fork of workers (SIGHUP to rspamd master), not
// `rspamadm control reload`
if sup != nil {
return nil, sup.Reload()
}
return nil, nil
})
t.Register("exec.relearn-spam", func(ctx context.Context, args map[string]any) (any, error) {
path, err := commands.ArgString(args, "file")
if err != nil {
return nil, err
}
data, err := os.ReadFile(path)
if err != nil {
return nil, err
}
r, err := commands.Run(ctx, commands.RunOptions{Stdin: data}, "rspamc", "learn_spam")
return nil, asError(r, err)
})
t.Register("exec.relearn-ham", func(ctx context.Context, args map[string]any) (any, error) {
path, err := commands.ArgString(args, "file")
if err != nil {
return nil, err
}
data, err := os.ReadFile(path)
if err != nil {
return nil, err
}
r, err := commands.Run(ctx, commands.RunOptions{Stdin: data}, "rspamc", "learn_ham")
return nil, asError(r, err)
})
return t
}
@@ -0,0 +1,87 @@
// Package services registers per-service command tables. The agent selects
// the right table at startup via MAILCOW_AGENT_SERVICE.
//
// A service "builder" receives a Supervisor for lifecycle commands; services
// that don't supervise a main process (currently just "host") pass nil and
// the generic lifecycle commands are skipped.
package services
import (
"context"
"fmt"
"time"
"github.com/mailcow/mailcow-dockerized/agent/internal/commands"
"github.com/mailcow/mailcow-dockerized/agent/internal/proc"
)
// Builder constructs a command table for a service. sup may be nil for
// services without a supervised main process.
type Builder func(sup *proc.Supervisor) *commands.Table
var registry = map[string]Builder{}
// Register installs a builder for a service name. Called from init() in each
// per-service file.
func Register(service string, b Builder) {
if _, dup := registry[service]; dup {
panic("services: duplicate registration for " + service)
}
registry[service] = b
}
// Build returns the table for service, or an error if no builder exists.
func Build(service string, sup *proc.Supervisor) (*commands.Table, error) {
b, ok := registry[service]
if !ok {
return nil, fmt.Errorf("services: unknown service %q (set MAILCOW_AGENT_SERVICE correctly)", service)
}
return b(sup), nil
}
// Known returns the list of registered service names (sorted-ish, depends on
// map iteration — for help output only).
func Known() []string {
out := make([]string, 0, len(registry))
for k := range registry {
out = append(out, k)
}
return out
}
// restartSettle is how long we wait after a Start to verify the new child
// didn't immediately crash. Gives the operator real "did the service come
// back up?" feedback instead of an instant OK that hides flapping services.
const restartSettle = 3 * time.Second
// addLifecycle wires reload/restart/stop/start onto t backed by sup. Services
// override these (e.g. postfix overrides reload to run `postfix reload`).
func addLifecycle(t *commands.Table, sup *proc.Supervisor) {
if sup == nil {
return
}
t.Register("reload", func(ctx context.Context, _ map[string]any) (any, error) {
return nil, sup.Reload()
})
t.Register("restart", func(ctx context.Context, _ map[string]any) (any, error) {
if err := sup.Restart(ctx); err != nil {
return nil, err
}
if err := sup.WaitStable(ctx, restartSettle); err != nil {
return nil, err
}
return map[string]any{"status": "restarted", "settled_ms": int(restartSettle / time.Millisecond)}, nil
})
t.Register("stop", func(ctx context.Context, _ map[string]any) (any, error) {
return nil, sup.Stop(ctx)
})
t.Register("start", func(ctx context.Context, _ map[string]any) (any, error) {
if err := sup.Start(); err != nil {
return nil, err
}
if err := sup.WaitStable(ctx, restartSettle); err != nil {
return nil, err
}
return map[string]any{"status": "started", "settled_ms": int(restartSettle / time.Millisecond)}, nil
})
}
@@ -0,0 +1,36 @@
package services
import (
"context"
"time"
"github.com/mailcow/mailcow-dockerized/agent/internal/commands"
"github.com/mailcow/mailcow-dockerized/agent/internal/proc"
)
func init() { Register("sogo", buildSogo) }
func sogoHealthProbe(ctx context.Context) error {
return probeHTTP(ctx, "http://127.0.0.1:20000/SOGo.index/", 3*time.Second)
}
func buildSogo(sup *proc.Supervisor) *commands.Table {
t := commands.New("sogo")
t.HealthProbe = sogoHealthProbe
addLifecycle(t, sup)
t.Register("exec.rename-user", func(ctx context.Context, args map[string]any) (any, error) {
oldName, err := commands.ArgString(args, "old")
if err != nil {
return nil, err
}
newName, err := commands.ArgString(args, "new")
if err != nil {
return nil, err
}
r, err := commands.Run(ctx, commands.RunOptions{}, "sogo-tool", "rename-user", oldName, newName)
return nil, asError(r, err)
})
return t
}
@@ -0,0 +1,26 @@
package services
import (
"context"
"time"
"github.com/mailcow/mailcow-dockerized/agent/internal/commands"
"github.com/mailcow/mailcow-dockerized/agent/internal/proc"
)
func init() { Register("unbound", buildUnbound) }
func unboundHealthProbe(ctx context.Context) error {
return probeShell(ctx, 3*time.Second, "dig", "+time=2", "+tries=1", "@127.0.0.1", "mailcow.email", "A")
}
func buildUnbound(sup *proc.Supervisor) *commands.Table {
t := commands.New("unbound")
t.HealthProbe = unboundHealthProbe
addLifecycle(t, sup)
t.Register("exec.flush-cache", func(ctx context.Context, _ map[string]any) (any, error) {
r, err := commands.Run(ctx, commands.RunOptions{}, "unbound-control", "flush_zone", ".")
return nil, asError(r, err)
})
return t
}